Update README.md

README.md

@@ -10,4 +10,7 @@ size_categories:
 We are thrilled to announce the HumanMOD dataset, which has been releases at our AMCIS 2023 paper. 
 
 Wang, Kanlun; Fu, Zhe; Zhou, Lina; and Zhang, Dongsong, "How Does User Engagement Support Content Moderation? A Deep Learning-based Comparative Study" (2023). AMCIS 2023 Proceedings. 3.
-https://aisel.aisnet.org/amcis2023/sig_aiaa/sig_aiaa/3
+https://aisel.aisnet.org/amcis2023/sig_aiaa/sig_aiaa/3
+
+Dataset Summary:
+The data collection was limited to public online communities to comply with the platform's privacy policy. Thus, the procedure did not require approval from the Institutional Review Board at the authors’ institution. We leveraged a Pushshift Reddit API to scrape posts from 40 subreddits daily across four different domains from Aug 24 to October 28, 2022, resulting in 104,674 posts. Since manual moderation is time-consuming, the efficiency of content moderation depends on several factors such as the type and volume of content, the complexity of the moderation rules, and the availability of human moderators. To enhance the ecological validity of the study findings, we used a PRAW API to perform another round of data collection of the collected posts 2 months later to validate whether the post content was moderated or not. Thereafter, we used a snowballing approach to collect the corresponding comments on all the posts. The metadata includes post content, post time, comment content, comment time, karma score, etc. In view that the volume of comments associated with each post varies widely from 0 to 500 (i.e., the maximum limit restricted by the API), and the distribution of the comments associated with each post is extremely right-skewed, we set a threshold for the minimum number of comments to 2 to facilitate the extraction of graph-based structural information. On the other hand, we also set an upper bound for the number of direct comments (i.e., the comments that directly respond to the original post) to 15. The final dataset, namely HumanMOD , is publicly available, which consists of 8,511 moderated posts and another 8,511 not moderated posts that were randomly selected from the remainder of the dataset. All the posts were commented on, with a total of 148,344 comments.